We propose the hierarchical Dirichlet process (HDP), a hierarchical, nonparametric, Bayesian model for clustering problems involving multiple groups of data. Such grouped clustering problems occur often in practice, e.g. in the problem of topic discovery in document corpora (Hoffman 1999, Blei et al 2003). Each group of data is modeled with a mixture, with the number of components being open-ended and inferred automatically by the model. Further, components can be shared across groups, allowing dependencies across groups to be modeled effectively as well as conferring generalization to new groups. HDPs are a principled solution to the grouped clustering problem, allowing a variety of different representations, and allowing for many possibilities for generalization. We report experimental results on three text corpora showing the effective and superior performance of the HDP over previous models.
Technical Report: Hierarchical Dirichlet processes. Teh, Jordan, Beal and Blei (2004). UC Berkeley Department of Statistics, TR 653. Can be obtained at: http://www.cs.berkeley.edu/~ywteh/research/npbayes